®
®
Richard Anderson
Consultant Production Code Generation
richard.anderson@mathworks.co.uk
© 2008 The MathWorks, Inc.
Algorithm Design and Code
Generation with Embedded MATLAB™
®
®
Agenda
Contribution of MATLAB functionality to a Model-Based
Design Workflow
What is the Embedded MATLAB functionality?
Demonstration
Converting a simple Median Filter to be C code generation
compatible.
Conclusion
2
®
®
But First a Quick Demonstration
3
®
®
Model-Based Design Workflow
Model algorithm using a high level modelling language
Execute system to test response and prove design
Generate code from block diagram
Either automatically
Or by hand
Test target implementation
If automatic code generation is used there is a single
source which can be referenced from each stage
4
®
®
Where does the algorithm originate?
Sometimes start with a blank sheet and develop it block
by block
This fits traditional Model-Based Design
Often originates from a prototype implementation written
in the m language.
How do you incorporate this?
How do you maintain traceability to original?
How can you reduce need to re-write algorithm?
5
®
®
Traditional Concept to Implementation:
Re-implement as you go down the level of abstraction
C/C++
C/C++code
code
M-code
M-code
Assembly
Assemblycode
code
HDL
HDL
Simulink
SimulinkModel
Model
Time-consuming, error-prone
Maintain multiple copies of the same algorithm
Need to fix and test across implementation boundaries
6
®
®
Better Concept to Implementation
One language
No multiple copies of source code
Elaborate M-code with real-world design constraints
Familiar development environment
Reuse of original test cases
Integrated visualization, analysis & debugging
Easy integration into Simulink environment
Automatic code generation
Path to embedded software (M to C)
7
®
®
What is the Embedded MATLAB
functionality?
Embedded MATLAB™ is the subset of the MATLAB®
language that can be compiled into code for embedded
implementation.
variable-sized data
objects java™
arrays
sparse
visualization
functions struct
Embedded
numeric fixed-point
complex
MATLAB
cellarrays
analysis
nested functions
varargin/varargout
8
®
®
Features of Embedded MATLAB
Extensive function library
270 MATLAB operators and functions
90 Fixed-Point Toolbox™ functions
Can automatically generate C code
EMLMEX – Generate C-Mex functions from M-code
EMLC – Generate C-Code directly from M-code
Reference M-files on the MATLAB path
Enables re-use of M-code in System Models
Enables partitioning of large Embedded MATLAB programs
Call custom C code directly with eml.ceval
Integration with Simulink and Stateflow®
Supports Frames, Structures, Data Type Override
9
®
®
What you will learn in this class
Make your MATLAB code compliant with the Embedded
MATLAB subset
Generate prototype C code automatically from the MATLAB
Desktop
Integrate your Embedded MATLAB compliant code into
Simulink models
Elaborate code to develop towards a production solution
Some tips for using Embedded MATLAB
10
®
®
Demonstration
Algorithm design with Embedded
MATLAB
Noise removal and
video enhancement
with
Median Filtering
11
®
®
Algorithm compliance with the Embedded
MATLAB subset
Typical steps to make MATLAB algorithms compliant
1. Remove instances of dynamic memory allocation
2. Lock down data types and sizes of variables
Assignment in function body
Assignment at compile time
3. Replace unsupported toolbox function calls with
Embedded MATLAB supported calls
4. Optimise algorithm for embedded implementation
12
®
®
Median Filter
Calculate Block Statistics
Replace Center Pixel or
Keep Center Pixel
Repeat for each Pixel
13
®
®
Median Filter
For each pixel in the image calculate the median value in a
surrounding neighborhood of the pixel.
Replace the starting pixel with the median value, if it’s value looks
skewed
Can affect all image pixels
Only performs well on low noise densities
Simple to implement
Computationally intensive
14
®
®
Summary (1)
Embedded MATLAB is the subset of MATLAB language
for embedded implementation
Supports 270 core MATLAB and 90 Fixed-Point Toolbox operators and
functions
Narrows concept-to-implementation design gap
Reduces algorithm translation, maintaining traceability
Embedded MATLAB Function block
Seamless integration of Embedded MATLAB compliant algorithm
into Simulink
Simulate the effect of your algorithm in the context of the whole
system
Automatically generate C code using Real-Time Workshop
15
®
®
Summary (2)
Embedded MATLAB in Simulink models provides
tools for better array manipulation
EMLC command
Automatically generate C source code from your compliant
MATLAB code directly from MATLAB desktop
EMLMEX command
Automatically generate C-Mex functions from compliant m-code
for algorithm acceleration
16
®
®
Products That Support Embedded MATLAB
Product
Functionality
Fixed-Point
Toolbox
Fixed-point functions and operators
emlmex command, which compiles M-files for accelerated execution
Simulink®
Embedded MATLAB Function block, which executes code in one or more Mfiles
Stateflow®
Embedded MATLAB functions that execute code in one or more M-files
SimEvents®
Attribute block that executes Embedded MATLAB code in one or more Mfiles
Real-Time
Workshop®
C code generation from MATLAB algorithms included in Simulink models,
Stateflow charts, or from the MATLAB command line (emlc command)
Simulink HDL
Coder™
Verilog and VHDL code generation from MATLAB algorithms included in
Simulink models and Stateflow charts
Simulink
Verification
and
Validation™
Develop designs and test cases mapped to requirements and measure test
coverage
Simulink
Design
Verifier™
Generate tests and prove model properties using formal methods
17
®
®
Bonus Tips for Efficient Code Generation
Examine algorithm to see if you can extract some sections
into Simulink.
Embedded MATLAB very good for array manipulation and
conditional execution.
Simulink better for complex computation
Examine use of colon (:) operator.
Remove it completely if operating on entire array/matrix
Unravel matrix operations and write out For loop you would expect
to see.
Inline functions were possible
18
®
®
Bonus Tips (Continued)
Embed Embedded MATLAB functions within Stateflow
and pass large data via ‘global’ chart level variables.
Reduces data copies
Can “encourage” pass-by-reference for sub-functions if
output is same variable as an input
Declare it as v = fcn(v) [Multiple inputs and outputs can be
supplied]
Make sure call to function assigns output to input variable
Control typing throughout your code to prevent
unnecessary up-casting.
i.e. Type literals (uint8(1)) when used with typed variables
19
®
®
Bonus Tips (Continued)
Turn off “Saturate on Integer Overflow” and build in
protection explicitly
Ensure %#eml is inserted in all m-files you expect to
generate C code or C-Mex files from
Ensures Embedded MATLAB M-lint checks are turned on
Improves error messages generated by emlc and emlmex
Use 1xN arrays rather than Nx1
20